AI Architect

Job highlights

• Comfortable reading and reviewing Java or TypeScript services. Production experience on AWS or GCP,containers and Kubernetes,and event-driven systems.
• Experience in supply chain,logistics or transportation.

Job Summary

At FourKites we have the opportunity to tackle complex challenges with real-world impacts. Whether its medical supplies from Cardinal Health or groceries for Walmart, the FourKites platform helps customers operate global supply chains that are efficient, agile and sustainable.

Join a team of curious problem solvers that celebrates differences, leads with empathy and values inclusivity.

As a Principal AI Architect, you will own the architecture underneath every FourKites AI agent: the harness they run in, the context they see, the models they call, and the evals that tell us whether they work. Our digital workers (Tracy for track and trace, Sam for supplier operations, Alan for appointment management, Cassie for customer service, Polly for document management, and more on the way) do real supply chain work on a shared platform, and you will set its direction. This is a hands-on individual contributor role based in Chennai (hybrid). Youll write code most weeks, lead design reviews across agent teams, and make the calls on model choice, context strategy and evaluation that every team builds on.

Responsibilities

• Define the reference architecture for FourKites agents: the agent loop, tool execution, state and checkpointing, retries and recovery, guardrails, permissions, and human-in-the-loop handoffs
• Design for long-running, multi-step work that spans many context windows and recovers cleanly when a tool, carrier API or model call fails
• Decide where each problem is best solved (frontier LLM, open-weight model, fine-tuned encoder, or plain code) and own model routing, cost-per-task and latency budgets across the fleet
• Integrate agents with FourKites event-driven platform (Kafka, microservices, Postgres) and with customer systems through APIs and MCP servers
• Set the standard for what goes into an agents context window (system prompts, tool definitions, retrieved documents, memory, prior turns) within an explicit token budget
• Build shared components for retrieval, compaction and summarization, sub-agent isolation, prompt caching and persistent memory so teams stop reinventing them
• Diagnose context failures (lost instructions, tool confusion, quality degradation as the window fills) and turn fixes into reusable harness patterns
• Build the offline eval framework: golden datasets from production traces, code-based checks, LLM-as-judge graders calibrated against human labels, and CI regression suites that gate every prompt, model or tool change
• Build online evaluation: scoring sampled production traffic, trace- and task-level quality metrics, drift detection, shadow and A/B rollouts, user feedback signals, and quality alerts
• Connect eval metrics to the business outcomes customers pay for, such as tracking accuracy, exceptions resolved without a human, and appointment fill rate
• Review designs and PRs for agent teams, write the architecture docs and playbooks they follow, and mentor staff and senior engineers
• Track the model and tooling landscape, run the experiments that matter, and give leadership clear build, buy or wait recommendations
• Partner with Product, Solutions and Customer Success to turn field failures into platform fixes

About the team

Our product and engineering teams are dedicated to providing the industrys best-in-class end-to-end supply chain visibility platform. The agent platform sits underneath every FourKites digital worker, and the teams building on it ship agents that track freight, chase suppliers, book dock appointments, answer customer questions and read shipping documents in production. Youll be the senior technical authority for how FourKites builds, ships and measures agents, working closely with agent teams in Chennai and with colleagues in the US and Europe.

Who you are

• Agents in production : youve architected and operated at least one LLM agent system with real users and real traffic (tool calling, multi-step workflows, failure handling, human escalation) and owned it after launch through incidents, cost and latency. Framework demos and POCs alone wont be a fit
• Deep ML knowledge of encoders and transformers : attention, positional encodings, encoder-only vs decoder-on…